tests/test_urcu_wfs.c

   1 /*
   2  * test_urcu_wfs.c
   3  *
   4  * Userspace RCU library - example RCU-based lock-free stack
   5  *
   6  * Copyright February 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  * Copyright February 2010 - Paolo Bonzini <pbonzini@redhat.com>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write to the Free Software Foundation, Inc.,
  21  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  22  */
  23
  24 #define _GNU_SOURCE
  25 #include "../config.h"
  26 #include <stdio.h>
  27 #include <pthread.h>
  28 #include <stdlib.h>
  29 #include <stdint.h>
  30 #include <stdbool.h>
  31 #include <string.h>
  32 #include <sys/types.h>
  33 #include <sys/wait.h>
  34 #include <unistd.h>
  35 #include <stdio.h>
  36 #include <assert.h>
  37 #include <sched.h>
  38 #include <errno.h>
  39
  40 #include <urcu/arch.h>
  41 #include <urcu/tls-compat.h>
  42
  43 #ifdef __linux__
  44 #include <syscall.h>
  45 #endif
  46
  47 /* hardcoded number of CPUs */
  48 #define NR_CPUS 16384
  49
  50 #if defined(_syscall0)
  51 _syscall0(pid_t, gettid)
  52 #elif defined(__NR_gettid)
  53 static inline pid_t gettid(void)
  54 {
  55         return syscall(__NR_gettid);
  56 }
  57 #else
  58 #warning "use pid as tid"
  59 static inline pid_t gettid(void)
  60 {
  61         return getpid();
  62 }
  63 #endif
  64
  65 #ifndef DYNAMIC_LINK_TEST
  66 #define _LGPL_SOURCE
  67 #endif
  68 #include <urcu.h>
  69 #include <urcu/wfstack.h>
  70
  71 /*
  72  * External synchronization used.
  73  */
  74 enum test_sync {
  75         TEST_SYNC_NONE = 0,
  76         TEST_SYNC_MUTEX,
  77 };
  78
  79 static enum test_sync test_sync;
  80
  81 static int test_force_sync;
  82
  83 static volatile int test_go, test_stop;
  84
  85 static unsigned long rduration;
  86
  87 static unsigned long duration;
  88
  89 /* read-side C.S. duration, in loops */
  90 static unsigned long wdelay;
  91
  92 static inline void loop_sleep(unsigned long loops)
  93 {
  94         while (loops-- != 0)
  95                 caa_cpu_relax();
  96 }
  97
  98 static int verbose_mode;
  99
 100 static int test_pop, test_pop_all;
 101
 102 #define printf_verbose(fmt, args...)            \
 103         do {                                    \
 104                 if (verbose_mode)               \
 105                         printf(fmt, ## args);   \
 106         } while (0)
 107
 108 static unsigned int cpu_affinities[NR_CPUS];
 109 static unsigned int next_aff = 0;
 110 static int use_affinity = 0;
 111
 112 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 113
 114 #ifndef HAVE_CPU_SET_T
 115 typedef unsigned long cpu_set_t;
 116 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 117 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 118 #endif
 119
 120 static void set_affinity(void)
 121 {
 122 #if HAVE_SCHED_SETAFFINITY
 123         cpu_set_t mask;
 124         int cpu, ret;
 125 #endif /* HAVE_SCHED_SETAFFINITY */
 126
 127         if (!use_affinity)
 128                 return;
 129
 130 #if HAVE_SCHED_SETAFFINITY
 131         ret = pthread_mutex_lock(&affinity_mutex);
 132         if (ret) {
 133                 perror("Error in pthread mutex lock");
 134                 exit(-1);
 135         }
 136         cpu = cpu_affinities[next_aff++];
 137         ret = pthread_mutex_unlock(&affinity_mutex);
 138         if (ret) {
 139                 perror("Error in pthread mutex unlock");
 140                 exit(-1);
 141         }
 142
 143         CPU_ZERO(&mask);
 144         CPU_SET(cpu, &mask);
 145 #if SCHED_SETAFFINITY_ARGS == 2
 146         sched_setaffinity(0, &mask);
 147 #else
 148         sched_setaffinity(0, sizeof(mask), &mask);
 149 #endif
 150 #endif /* HAVE_SCHED_SETAFFINITY */
 151 }
 152
 153 /*
 154  * returns 0 if test should end.
 155  */
 156 static int test_duration_dequeue(void)
 157 {
 158         return !test_stop;
 159 }
 160
 161 static int test_duration_enqueue(void)
 162 {
 163         return !test_stop;
 164 }
 165
 166 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 167 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 168
 169 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 170 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 171
 172 static unsigned int nr_enqueuers;
 173 static unsigned int nr_dequeuers;
 174
 175 static struct cds_wfs_stack s;
 176
 177 static void *thr_enqueuer(void *_count)
 178 {
 179         unsigned long long *count = _count;
 180
 181         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 182                         "enqueuer", (unsigned long) pthread_self(),
 183                         (unsigned long) gettid());
 184
 185         set_affinity();
 186
 187         while (!test_go)
 188         {
 189         }
 190         cmm_smp_mb();
 191
 192         for (;;) {
 193                 struct cds_wfs_node *node = malloc(sizeof(*node));
 194                 if (!node)
 195                         goto fail;
 196                 cds_wfs_node_init(node);
 197                 cds_wfs_push(&s, node);
 198                 URCU_TLS(nr_successful_enqueues)++;
 199
 200                 if (caa_unlikely(wdelay))
 201                         loop_sleep(wdelay);
 202 fail:
 203                 URCU_TLS(nr_enqueues)++;
 204                 if (caa_unlikely(!test_duration_enqueue()))
 205                         break;
 206         }
 207
 208         count[0] = URCU_TLS(nr_enqueues);
 209         count[1] = URCU_TLS(nr_successful_enqueues);
 210         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 211                        "enqueues %llu successful_enqueues %llu\n",
 212                        pthread_self(),
 213                         (unsigned long) gettid(),
 214                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 215         return ((void*)1);
 216
 217 }
 218
 219 static void do_test_pop(enum test_sync sync)
 220 {
 221         struct cds_wfs_node *node;
 222
 223         if (sync == TEST_SYNC_MUTEX)
 224                 cds_wfs_pop_lock(&s);
 225         node = __cds_wfs_pop_blocking(&s);
 226         if (sync == TEST_SYNC_MUTEX)
 227                 cds_wfs_pop_unlock(&s);
 228
 229         if (node) {
 230                 free(node);
 231                 URCU_TLS(nr_successful_dequeues)++;
 232         }
 233         URCU_TLS(nr_dequeues)++;
 234 }
 235
 236 static void do_test_pop_all(enum test_sync sync)
 237 {
 238         struct cds_wfs_head *head;
 239         struct cds_wfs_node *node, *n;
 240
 241         if (sync == TEST_SYNC_MUTEX)
 242                 cds_wfs_pop_lock(&s);
 243         head = __cds_wfs_pop_all(&s);
 244         if (sync == TEST_SYNC_MUTEX)
 245                 cds_wfs_pop_unlock(&s);
 246
 247         cds_wfs_for_each_blocking_safe(head, node, n) {
 248                 free(node);
 249                 URCU_TLS(nr_successful_dequeues)++;
 250                 URCU_TLS(nr_dequeues)++;
 251         }
 252 }
 253
 254 static void *thr_dequeuer(void *_count)
 255 {
 256         unsigned long long *count = _count;
 257         unsigned int counter;
 258
 259         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 260                         "dequeuer", (unsigned long) pthread_self(),
 261                         (unsigned long) gettid());
 262
 263         set_affinity();
 264
 265         while (!test_go)
 266         {
 267         }
 268         cmm_smp_mb();
 269
 270         assert(test_pop || test_pop_all);
 271
 272         for (;;) {
 273                 if (test_pop && test_pop_all) {
 274                         if (counter & 1)
 275                                 do_test_pop(test_sync);
 276                         else
 277                                 do_test_pop_all(test_sync);
 278                         counter++;
 279                 } else {
 280                         if (test_pop)
 281                                 do_test_pop(test_sync);
 282                         else
 283                                 do_test_pop_all(test_sync);
 284                 }
 285
 286                 if (caa_unlikely(!test_duration_dequeue()))
 287                         break;
 288                 if (caa_unlikely(rduration))
 289                         loop_sleep(rduration);
 290         }
 291
 292         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 293                        "dequeues %llu, successful_dequeues %llu\n",
 294                        pthread_self(),
 295                         (unsigned long) gettid(),
 296                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 297         count[0] = URCU_TLS(nr_dequeues);
 298         count[1] = URCU_TLS(nr_successful_dequeues);
 299         return ((void*)2);
 300 }
 301
 302 static void test_end(struct cds_wfs_stack *s, unsigned long long *nr_dequeues)
 303 {
 304         struct cds_wfs_node *node;
 305
 306         do {
 307                 node = cds_wfs_pop_blocking(s);
 308                 if (node) {
 309                         free(node);
 310                         (*nr_dequeues)++;
 311                 }
 312         } while (node);
 313 }
 314
 315 static void show_usage(int argc, char **argv)
 316 {
 317         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 318         printf(" [-d delay] (enqueuer period (in loops))");
 319         printf(" [-c duration] (dequeuer period (in loops))");
 320         printf(" [-v] (verbose output)");
 321         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 322         printf(" [-p] (test pop)");
 323         printf(" [-P] (test pop_all, enabled by default)");
 324         printf(" [-M] (use mutex external synchronization)");
 325         printf("      Note: default: no external synchronization used.");
 326         printf(" [-f] (force user-provided synchronization)");
 327         printf("\n");
 328 }
 329
 330 int main(int argc, char **argv)
 331 {
 332         int err;
 333         pthread_t *tid_enqueuer, *tid_dequeuer;
 334         void *tret;
 335         unsigned long long *count_enqueuer, *count_dequeuer;
 336         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 337         unsigned long long tot_successful_enqueues = 0,
 338                            tot_successful_dequeues = 0;
 339         unsigned long long end_dequeues = 0;
 340         int i, a;
 341
 342         if (argc < 4) {
 343                 show_usage(argc, argv);
 344                 return -1;
 345         }
 346
 347         err = sscanf(argv[1], "%u", &nr_dequeuers);
 348         if (err != 1) {
 349                 show_usage(argc, argv);
 350                 return -1;
 351         }
 352
 353         err = sscanf(argv[2], "%u", &nr_enqueuers);
 354         if (err != 1) {
 355                 show_usage(argc, argv);
 356                 return -1;
 357         }
 358
 359         err = sscanf(argv[3], "%lu", &duration);
 360         if (err != 1) {
 361                 show_usage(argc, argv);
 362                 return -1;
 363         }
 364
 365         for (i = 4; i < argc; i++) {
 366                 if (argv[i][0] != '-')
 367                         continue;
 368                 switch (argv[i][1]) {
 369                 case 'a':
 370                         if (argc < i + 2) {
 371                                 show_usage(argc, argv);
 372                                 return -1;
 373                         }
 374                         a = atoi(argv[++i]);
 375                         cpu_affinities[next_aff++] = a;
 376                         use_affinity = 1;
 377                         printf_verbose("Adding CPU %d affinity\n", a);
 378                         break;
 379                 case 'c':
 380                         if (argc < i + 2) {
 381                                 show_usage(argc, argv);
 382                                 return -1;
 383                         }
 384                         rduration = atol(argv[++i]);
 385                         break;
 386                 case 'd':
 387                         if (argc < i + 2) {
 388                                 show_usage(argc, argv);
 389                                 return -1;
 390                         }
 391                         wdelay = atol(argv[++i]);
 392                         break;
 393                 case 'v':
 394                         verbose_mode = 1;
 395                         break;
 396                 case 'p':
 397                         test_pop = 1;
 398                         break;
 399                 case 'P':
 400                         test_pop_all = 1;
 401                         break;
 402                 case 'M':
 403                         test_sync = TEST_SYNC_MUTEX;
 404                         break;
 405                 case 'f':
 406                         test_force_sync = 1;
 407                         break;
 408                 }
 409         }
 410
 411         /* activate pop_all test by default */
 412         if (!test_pop && !test_pop_all)
 413                 test_pop_all = 1;
 414
 415         if (test_sync == TEST_SYNC_NONE && nr_dequeuers > 1 && test_pop) {
 416                 if (test_force_sync) {
 417                         fprintf(stderr, "[WARNING] Using pop concurrently "
 418                                 "with other pop or pop_all without external "
 419                                 "synchronization. Expect run-time failure.\n");
 420                 } else {
 421                         printf("Enforcing mutex synchronization\n");
 422                         test_sync = TEST_SYNC_MUTEX;
 423                 }
 424         }
 425
 426         printf_verbose("running test for %lu seconds, %u enqueuers, "
 427                        "%u dequeuers.\n",
 428                        duration, nr_enqueuers, nr_dequeuers);
 429         if (test_pop)
 430                 printf_verbose("pop test activated.\n");
 431         if (test_pop_all)
 432                 printf_verbose("pop_all test activated.\n");
 433         if (test_sync == TEST_SYNC_MUTEX)
 434                 printf_verbose("External sync: mutex.\n");
 435         else
 436                 printf_verbose("External sync: none.\n");
 437         printf_verbose("Writer delay : %lu loops.\n", rduration);
 438         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 439         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 440                         "main", (unsigned long) pthread_self(),
 441                         (unsigned long) gettid());
 442
 443         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 444         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 445         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 446         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 447         cds_wfs_init(&s);
 448
 449         next_aff = 0;
 450
 451         for (i = 0; i < nr_enqueuers; i++) {
 452                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 453                                      &count_enqueuer[2 * i]);
 454                 if (err != 0)
 455                         exit(1);
 456         }
 457         for (i = 0; i < nr_dequeuers; i++) {
 458                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 459                                      &count_dequeuer[2 * i]);
 460                 if (err != 0)
 461                         exit(1);
 462         }
 463
 464         cmm_smp_mb();
 465
 466         test_go = 1;
 467
 468         for (i = 0; i < duration; i++) {
 469                 sleep(1);
 470                 if (verbose_mode)
 471                         write (1, ".", 1);
 472         }
 473
 474         test_stop = 1;
 475
 476         for (i = 0; i < nr_enqueuers; i++) {
 477                 err = pthread_join(tid_enqueuer[i], &tret);
 478                 if (err != 0)
 479                         exit(1);
 480                 tot_enqueues += count_enqueuer[2 * i];
 481                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 482         }
 483         for (i = 0; i < nr_dequeuers; i++) {
 484                 err = pthread_join(tid_dequeuer[i], &tret);
 485                 if (err != 0)
 486                         exit(1);
 487                 tot_dequeues += count_dequeuer[2 * i];
 488                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 489         }
 490
 491         test_end(&s, &end_dequeues);
 492
 493         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 494                        tot_enqueues, tot_dequeues);
 495         printf_verbose("total number of successful enqueues : %llu, "
 496                        "successful dequeues %llu\n",
 497                        tot_successful_enqueues, tot_successful_dequeues);
 498         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 499                 "nr_dequeuers %3u "
 500                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 501                 "successful enqueues %12llu successful dequeues %12llu "
 502                 "end_dequeues %llu nr_ops %12llu\n",
 503                 argv[0], duration, nr_enqueuers, wdelay,
 504                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 505                 tot_successful_enqueues,
 506                 tot_successful_dequeues, end_dequeues,
 507                 tot_enqueues + tot_dequeues);
 508         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 509                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 510                        "succ. dequeues + end dequeues %llu.\n",
 511                        tot_successful_enqueues,
 512                        tot_successful_dequeues + end_dequeues);
 513
 514         free(count_enqueuer);
 515         free(count_dequeuer);
 516         free(tid_enqueuer);
 517         free(tid_dequeuer);
 518         return 0;
 519 }